Kategorien durchsuchen
Entdecken
Fiverr Pro
Deutsch
$
USD
I will train a custom yolo model for object detection, tracking and counting
Luong D
Über diesen Service
Need a model that reliably detects your objects on real images, not only in a notebook? I train custom YOLO models (Ultralytics YOLOv8/YOLO11: detection, segmentation or pose) on your dataset.
Need tracking or counting too? I add object tracking (ByteTrack/BoT-SORT), line-crossing counts and zone alerts on video.
What I have built:
- Real-time fall detection from RTSP cameras (YOLO Pose, tracking, ONNX, FastAPI, Docker). 83% F1 on a 100-video benchmark, Top 10 at AI20K (Vingroup x VinUni).
- PPE safety detection: cleaned a 7,000-image multi-source dataset and compared 10+ YOLO methods. mAP50 improved from 0.674 to 0.753 on an 808-image test set.
What you get:
- Trained weights (.pt, plus ONNX in Premium)
- mAP50, mAP50-95, precision, recall, per-class results and confusion matrix
- A Python inference script for images, video or webcam
- A short report on what works, what fails and how to improve it
- A clean README so you can retrain later
Process: send your dataset (YOLO, COCO, VOC or Roboflow), I check it for label problems and leakage, train, evaluate on a held-out test set, and deliver.
Data not labeled yet? Message me first.
Lerne Luong D kennen
Luong D
AI Engineer for Computer Vision and LLM Apps
- AusVietnam
- Mitglied seitJuli 2021
- ⌀ Antwortzeit1 Stunde
Sprachen
Vietnamesisch, Englisch
I build AI systems that work in the real world. My recent projects include SYL GuardianCam, a real-time fall detection system (YOLO Pose, FastAPI, Docker) with 83% F1 that placed Top 10 at AI20K, and a Medical GraphRAG system that reduces LLM hallucinations. I can train custom YOLO models, clean datasets, build RAG chatbots and AI agents, and deploy them with FastAPI and Docker. Every delivery includes clean code, a README and clear metrics.
FAQ
What format should my dataset be in?
YOLO TXT, COCO JSON, Pascal VOC or a Roboflow export. I can convert between them, so any of these is fine.
Do I need a GPU?
No. I train on my own setup and deliver weights that also run on CPU. Premium includes an ONNX export for faster CPU or edge inference.
Can you guarantee a specific mAP?
Not before seeing the data. Accuracy depends mostly on data quality. I report real metrics on a held-out test set and explain exactly what would improve them.
Can you count or track objects in video?
Yes. I add tracking (ByteTrack or BoT-SORT) so each object keeps an ID, then count line crossings or time spent in a zone. You get the counts as CSV/JSON plus an annotated video. My fall-detection system already uses multi-person tracking on live camera streams.
Will it run on CPU, Jetson or Raspberry Pi?
Usually yes, with a small model (YOLO11n/s) exported to ONNX. Speed depends on the device and image size, so I don't promise FPS up front. I report the measured CPU speed and give you a test script to measure FPS on your Jetson or Pi before you accept delivery.
Do you sign an NDA and delete my data?
Yes. I can sign a reasonable NDA. Your images and labels are used only for your order, never shared or reused for anything else, and I delete them after delivery if you ask.
